Batch phpcs into two invocations: one for new files, one for old

[sirbrillig]

Fixes #115

Summary

Alternative to #110, which batches all phpcs calls into a single invocation.

This approach uses two separate phpcs batch calls instead of one combined call:

1. First batch: Run phpcs on all new (modified) file versions at once
2. Second batch: Run phpcs on all old (unmodified) file versions at once — but only for files that have new phpcs messages

Key difference from #110

The single-batch approach (#110) pre-determines which old files need scanning before the new-file results are known, which means it sometimes scans old versions of files that turn out to have no new messages (wasted work).

This two-batch approach waits until the new-file results are in before deciding which old versions to scan, preserving the existing per-file optimization of skipping old-version scans for clean files.

Trade-offs vs #110

• Simpler: Each batch method returns array<string,string> instead of array{new:..., old:...}
• Smarter: Never scans old versions of files with no new messages
• Two startup costs instead of one when all files have messages (minor overhead)

Performance

Measured with benchmark.sh on 10 staged PHP files (PSR2 standard, 10 runs):

	Mean	Relative
trunk: 20 phpcs calls	6.94 s ± 0.61 s	1.00×
two-batch (this branch): 2 phpcs calls	3.57 s ± 0.29 s	1.95× faster